An algorithm with preferences: from zero to one variable — LessWrong